When does Privileged Information Explain Away Label Noise?
This work provides insights into PI's role in label noise, which is a common problem in machine learning, though it appears incremental with enhancements to existing methods.
The study investigated how privileged information (PI) helps address label noise, finding it's most effective when it distinguishes clean from noisy data and allows memorization of noisy examples, but performance drops if PI is too predictive of the target label.
Leveraging privileged information (PI), or features available during training but not at test time, has recently been shown to be an effective method for addressing label noise. However, the reasons for its effectiveness are not well understood. In this study, we investigate the role played by different properties of the PI in explaining away label noise. Through experiments on multiple datasets with real PI (CIFAR-N/H) and a new large-scale benchmark ImageNet-PI, we find that PI is most helpful when it allows networks to easily distinguish clean from noisy data, while enabling a learning shortcut to memorize the noisy examples. Interestingly, when PI becomes too predictive of the target label, PI methods often perform worse than their no-PI baselines. Based on these findings, we propose several enhancements to the state-of-the-art PI methods and demonstrate the potential of PI as a means of tackling label noise. Finally, we show how we can easily combine the resulting PI approaches with existing no-PI techniques designed to deal with label noise.